Evaluation of adaptive algorithms for detection and classification of fluorescent aerosols in the atmosphere

Abstract

Photon counting technologies are developed and could be used in the future to measure the return from laser induced fluorescence. Currently, the spectral detection of light emitted by fluorescing aerosols is performed with ICCD, Intensified Charge Coupled Device. The signal to noise ratio of ICCD devices is smaller by a factor of √2compared to photon counting devices having the same sensitivity. We studied the impact of this difference of signal to noise ratio on the capability of multivariate detection and classification algorithms to operate on various conditions. Signal simulations have been performed to obtain ROC (Receiver Operation Characteristics) Curves and Confusion Matrix to obtain the detection performance and the ability of algorithms to discriminate a potential source from another. Two detection algorithms are used, the Integrated Laser Induced Fluorescence(ILIF) and the Matched Filter. For the classification, three algorithms are used, the Adaptive Matched Filter (AMF), the Adaptive Coherent Estimator (ACE) and the Adaptive Least Squares (ALS). The best algorithm for detection is the AMF using the signature of the material present in a cloud, the ILIF detector performs very well. For the classification, the three algorithms are surprisingly giving the same results for the same data. The classification performs better if the distance between the signatures recorded in a database is important. The performance of the detector and of the classificator improves with an increase of the signal to noise ratio and is consistently and significantly better for the photon counting compared to ICCD.